Operator perspectives.

We didn't set out to build a voice AI. We set out to stop losing customers.

At our Subway locations in San Diego, we kept noticing the same thing: phones ringing during the lunch rush with nobody picking up. Not because we didn't care — because the team was buried making sandwiches, the line was out the door, and the phone rang at exactly the wrong moment. Every time it went unanswered, we knew what happened next. That customer ordered from somewhere else.

At any given moment, roughly 30% of inbound calls to Subway locations go unanswered. That's not a staffing problem. That's a structural problem.

The math that kept us up at night

Think about what an unanswered call actually costs. A regular sandwich order is $15–25. A catering inquiry is $200–600. If your store gets 50 calls a day and misses 15 of them, you're leaking thousands of dollars a week — quietly, invisibly, with no alert and no record.

We tried the obvious fixes first. We hired more staff. We tried call-routing software. We posted signs asking customers to leave voicemails. None of it worked. Voicemails don't get called back. Routing software dumps callers into menus that frustrate them. And hiring for phone coverage is expensive and unreliable — people call in sick, turnover is high, and the lunch rush hits everyone at once.

What actually worked

We started exploring AI voice agents about two years ago — not as a gimmick, but as genuine operational infrastructure. The question we asked wasn't "can AI take a phone call?" It was: "can AI handle this well enough that a customer doesn't feel like they're talking to a robot?"

The answer, after a lot of iteration, was yes — but only if you build it right. The agent needs to know your menu cold. It needs to handle modifications, substitutions, and catering lead times without hesitation. It needs to upsell naturally, not robotically. And critically: it needs to know when to hand off to a human and do it gracefully.

Maya — the voice agent we built — now picks up on the first ring at every one of our stores. She takes sandwich orders, handles catering inquiries, answers menu questions, and warm-transfers to the manager the moment a customer asks for a person. Call volume captured went from roughly 70% to 100%. Catering bookings went up. And our team stopped getting interrupted during service.

What we learned building it

A few things surprised us along the way. First: customers adapt fast. Within a few calls, they're ordering from Maya like they've done it for years. The friction we worried about didn't materialize — as long as the voice felt natural and the agent actually knew what it was doing.

Second: the upsell matters more than we expected. Maya pitches avocado, double meat, and the catering bundle naturally — not as a scripted prompt but as part of the conversation. Average order value went up without us doing anything different on the menu side.

Third: after-hours is where the real upside lives. Catering customers often call at 8pm to book for the next morning. Before Maya, those calls went to voicemail and got lost. Now they get captured, confirmed, and logged — every time.

If you're running more than one location and you're still relying on your team to answer every phone call, you're leaving money on the table. We built Maya because we needed her. Now she runs in every one of our stores — and we've made her available to other operators who are tired of watching calls go unanswered.

Catering customers are the most valuable customers in your Subway business. They order in bulk, they tip well, they're predictable, and when they find a store they trust, they come back repeatedly. They're also the customers most operators completely fail to follow up with.

I've been running Subway locations in San Diego for years. When we started auditing our catering customer database, we found something that genuinely surprised us: a huge percentage of our best past catering customers had simply stopped ordering — not because they were unhappy, but because nobody had followed up. They'd placed one or two big orders, had a fine experience, and then quietly drifted to whoever happened to answer their next call.

The average catering order at a Subway is $300–600. Most operators have dozens of these customers sitting dormant in their data — and they're doing nothing about it.

Why catering customers churn quietly

Regular customers come back on their own. They walk past your store, they're hungry, they come in. Catering customers are different. They only order when they have an event — a lunch meeting, a team training, a school event. If you're not in their mind when that moment arrives, they'll search for whoever comes up first, or call wherever they called last time.

This is a retention problem disguised as a new customer problem. Operators spend money on ads trying to find new catering accounts when their existing database is full of people who already trust them. The reactivation opportunity is sitting there unused.

What a reactivation sequence actually looks like

We built our reactivation system in stages. Here's the sequence that works best for us:

Day 15: A short thank-you SMS goes out after the original order. "Hope the lunch was a hit — we'd love a Google review if you have a minute." This builds the relationship and often generates a review while the experience is fresh.

Day 30: If no new order has come in, a reactivation SMS goes out. Personal, specific. "Hey [name] — it's been about a month since your office order. Got another meeting coming up? We can take care of it." No blast, no promo code — just a human-feeling nudge.

Day 45: Maya makes an outbound call. She references the customer's last order, asks about upcoming events, and offers to book on the spot. This is where the real magic happens — a personalized voice call that most operators would never have the bandwidth to make manually.

Day 60: Final email with a light offer — a free cookie tray with the next catering order, or a bundle deal. After this, if there's no response, the customer is marked as churned and removed from the active sequence.

The numbers that changed how we think about this

When we started running this sequence properly, we were surprised by how many customers responded — not because the messages were aggressive, but because they were timely and personal. Office managers are busy. They don't plan catering weeks in advance. When a reminder lands at the right moment, it converts.

If you have a catering customer list and you're not running a reactivation sequence, start with the 30-day SMS. It takes five minutes to set up and the ROI is immediate. The $300 customer is already in your database. You just need to go get them.

Ask most Subway operators about Google reviews and they'll say something like: "Yeah, we try to respond when we remember." That's leaving serious money on the table.

Google reviews aren't just a reputation signal — they're a local SEO driver. The number of reviews you have, how recently they were posted, and how quickly you respond to them all factor into how Google ranks your location in local search. When someone searches "Subway near me" or "sandwich delivery San Diego," the stores that show up first are almost always the ones with the most reviews and the most consistent engagement.

A store that goes from 3.8 stars to 4.4 stars with 200+ reviews doesn't just look better — it ranks higher, gets more clicks, and sees more walk-ins. The reviews are the product.

The three levers most operators miss

Asking for reviews systematically. The single best predictor of review volume is whether you ask. Happy customers don't leave reviews unprompted — they're busy, they forget. A simple SMS sent 24 hours after a catering order or in-store visit converts at a surprisingly high rate. Most operators never send it.

Responding to every review within an hour. Google's algorithm rewards response rate and response speed. A store that replies to 95% of reviews within an hour signals to Google that it's actively managed and customer-focused. That signal translates to ranking. Most operators respond to maybe 20% of reviews, days later, with copy-pasted text.

Taking 1–3 star reviews seriously. A negative review left unaddressed is a conversion killer for anyone reading your profile. A negative review with a thoughtful, prompt response from the owner turns a liability into proof of character. "We're sorry this happened, here's our number, we want to make it right" — that message isn't just for the unhappy customer. It's for the next 50 people who read that review before deciding where to order.

What we automated and what we kept human

At our stores, we auto-reply to every review within the hour using AI — but the responses are personalized, brand-voiced, and genuinely engaging. They reference what the customer said, match the tone of the review, and feel like a real person wrote them.

What we kept human: when a 1 or 2 star review comes in, our manager gets an SMS alert immediately. The auto-reply goes out fast to show we're paying attention — but a human follows up directly when needed. That combination — speed from AI, judgment from humans — is what works in practice.

We went from a 3.9 average across our stores to a 4.6 average in about six months. Not by doing anything dramatic — just by asking for reviews consistently, replying to all of them fast, and actually paying attention to the bad ones. If you're not doing those three things, start today. The reviews compound. Your competitors are already behind you before you've written a word.